Claude Skills by fagemx

Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.

Use when a concept needs a multi-file design dossier (overview, canonical form, schemas, rules, APIs, slices, demo paths) before it's stable enough for project-plan task decomposition. Not for single-file RFCs, ADRs, or already-stable scopes ready for engineering.

Deep code review and quality analysis for the edda Rust project

Review merged code, commit ranges, or codebase paths for correctness, security, architecture, and test risk; use pr-review for open PRs

Complete pre-commit workflow - run quality checks (format, lint, test) and validate/create conventional commit messages

Clean handoff when finishing multi-agent work — summarize, decide, unclaim

Use when coordinating two or more sessions on parallel implementation work as the controller — assigning bundles, briefing workers and a verifier, adjudicating mid-flight, and closing with dual review

Detect changes affecting peers and send coordination requests

Use when auditing multi-session coordination or reviewing a delivery candidate before PR or merge

Sync with peers before starting multi-agent work — see claims, recorded decisions, suggest scope

Dispatch tasks to Karvi with correct runtime, provider, and model configuration

Use when refilling the Fleet feature queue — decompose one signed-off goal or one section of an existing PLAN into atomic, independently-mergeable fleet:pending issues. Recon before decompose; never write implementation code.

Use when a controller must coordinate two or more agent sessions for parallel review, issue discovery, implementation, verification, dependency conflicts, or unattended progress that must survive context loss or session replacement.

Use when driving a fleet PR to LGTM hands-off — a deterministic bash driver alternates an independent fleet-review (gate) and an independent fix pass until LGTM or a round cap, posting each verdict to the PR, then stops for the operator to merge. Never merges (GATE-01). Orchestrates fleet-review + a fix pass; reads conventions from the repo's own CLAUDE.md / AGENTS.md.

Use when gating a fleet PR before merge — independently (fork) verify the repo's gates on the verification ladder (READ receipts and exact-head CI; RAN only focused checks), adversarially review the diff against the linked issue's doneWhen and repo conventions, post the verdict as a PR comment, and stop. Never fixes, never merges (GATE-01). Reads conventions from the repo's own CLAUDE.md / AGENTS.md.

Use when coordinating engineering work across mixed agent runtimes on this machine — pi terminal sessions as concurrent workers, a separate Claude session as independent verifier, this session as controller — or when you can only reach a target agent via terminal paste, Run-button relay, or cross-session SendMessage (dispatch, monitor, freeze, verdict, merge).

Use when running one lane of the Fleet execution loop — pick a signed-off fleet:ready issue, implement it in an isolated worktree with TDD, open a PR, and stop. Reads conventions from the repo's own CLAUDE.md / AGENTS.md.

Check if codebase supports a planned task — what's ready, what's close, what's missing

Periodic codebase health assessment — 5 dimensions, scorecard, trend tracking

Dual-agent PR intelligence scan — discover contribution opportunities from any repo's PR history

Continue working on GitHub issue from conversation context

Consolidate GitHub issue discussion into clean body for handoff

Create GitHub issues from conversation context (general, bug reports, or feature requests)

Use when the ready-issue queue is low or a delivery wave is closing, when a defect/finding surfaces mid-session (runtime failure, review discovery, repeated manual step), or when deciding whether an observation deserves a GitHub issue. Also use when authoring any issue body.

End-to-end issue pipeline: plan → implement → review → merge. Dispatches parallel sub-agents with worktree isolation. Usage: /issue-pipeline 566 567 568 [--skip-plan] [--no-merge]

Start working on GitHub issue with deep-dive workflow (research, innovate, plan phases)

Scan codebase, discover work, propose well-scoped issues

Use when two or more ready issues are about to be delivered serially, when authoring or restructuring a conduct wave plan, or when deciding whether concurrent implementation lanes are safe on this workstation (worktree, build lane, review dispatch, merge ordering).

Find the most unblocking next step given current codebase state and goals

Decompose planning docs into GitHub issues — extract gaps, deduplicate, confirm with human, batch create

Validate a plan for gaps by cross-referencing against actual code, call chains, constraints, and test edge cases

Monitor PR CI pipeline, auto-fix issues, and loop until all checks pass

Iteratively review PR, post comment, fix issues, and re-review until LGTM

PR review: concise, direct, actionable

Generate a structured planning pack (tracks, tasks, contracts, validation) for any project

Core architectural and code quality principles that guide all development decisions in the edda project

PR lifecycle management - create PRs with proper commits, merge with validation, and manage PR comments

Systematic scan → classify → batch fix → add automation to prevent regression

Create command aliases for skills with arguments

Create high-quality Claude Code skills following edda patterns

Test a Claude Code skill — verify it runs, references exist, output is sane

Review architecture specs, RFCs, API docs, design docs, decision-system docs, or any multi-file technical spec stack using three explicit reading modes: (A) quick overview, (B) strict spec audit, and (C) product/architecture critique. Use when the user asks to 看 spec, review docs, 掃一遍架構, 幫我抓矛盾, 做 spec 審查, or wants structured feedback on boundaries, type consistency, lifecycle ownership, API seams, product thickness, or whether a spec feels complete vs actually useful.

Technical debt management - scan Rust codebase for bad smells and create tracking issues

Comprehensive testing patterns and anti-patterns for writing and reviewing tests

Analyze complex problems structurally — map actors, find root causes, create tracking issue with parallel task decomposition

Convert broad goals, long-running projects, learning plans, research tasks, repo modernization, product builds, writing projects, operations work, portfolio prep, or job-search preparation into a proactive verifiable active goal run operating loop. Use when the user wants goal.md/state.md/backlog.md/verifier.md, external stop conditions, sprint/cycle planning, durable progress across long tasks, skill/tool routing, command prompts, autonomy rules, or a reusable method that prevents passive on...

Multi-phase AI plan orchestration via edda-conductor. Use when: (1) user wants to run a multi-phase plan, (2) checking conductor status or progress, (3) handling blocked/failed phases (retry, skip, abort), (4) generating plan.yaml from natural language. NOT for: single-task execution (use coding-agent), decision recording (use edda decide), session context (handled by edda bridge).

Post-task notification and draft approval via edda (decision memory). Use when: (1) a coding agent finishes and you need to report results, (2) checking for pending edda drafts that need approval, (3) user says 'approve', 'reject', or asks about draft status. NOT for: recording decisions (use `edda decide` directly), querying past decisions (use `edda ask`), or session context (handled automatically by edda bridge).

Use when a controller must coordinate two or more agent sessions for parallel review, issue discovery, implementation, verification, dependency conflicts, or unattended progress that must survive context loss or session replacement.